27 stories
Show: Summaries / Headlines only
IBM researchers discarded runs that failed error checks on 49 physical qubits of a Heron processor, then applied error cancellation to the remaining noise. They inferred a sampling cost of 1,359 at six time steps, versus 85,545 without the checks.
A Sandia-led team scored Google, IBM, and Quantinuum processors and a logical-qubit machine on one scale. Helios-1 led at 1,504 without discarded runs, against about 250 million that the authors calculated for factoring RSA-2048.
Harvard researchers protected a silicon-vacancy qubit in diamond with a continuous acoustic drive in place of microwave pulses. They ran the experiment on a test chip without the phononic cavities meant to link qubits in a future network.
Five physicists calculated that an Andreev qubit in a superposition should produce an oscillating supercurrent with no applied voltage. In simulations described in their preprint, they read one test state from that current with infidelities of 0.15% to 0.48%.
In a preprint, Qarakal's researchers counted the physical qubits for one merge operation and reached the tenfold ratio by extrapolating beyond the code distances they simulated. The company has not reported a Pangaea chip.
About 0.5% of gates ended in a detected erasure and at most 0.12% in an undetected error. The authors projected the error-correction advantage from simulations, assuming perfect erasure checks and leaving out measurement errors.
A 70-million-transistor chip at 4 kelvin generated every control pulse for HRL's silicon spin qubits. The team used up to seven of its 18 physical qubits, on codes that correct one type of error or only detect errors.
IBM and University of Chicago researchers claimed advantage with a certified fidelity bound on 97 physical qubits. Qedma and Algorithmiq reported regimes in which the classical methods they tested became unreliable or disagreed. None of the preprints is peer reviewed.
The team sent the largest fragments of small molten-salt clusters to an IBM processor, using circuits of up to 66 qubits plus ancilla qubits. The authors traced the dominant remaining error to the classical step that split the clusters.
The authors called the properties predictability, typicality, robustness, verifiability, and usefulness. They wrote that Shor's algorithm satisfies their criteria and proved that some advantages cannot be predicted with classical resources alone.
